Nonlinear supervised dimensionality reduction via smooth regular embeddings

作者：

Highlights：

• A nonlinear supervised dimensionality reduction method is proposed.

• The learning objective is based on recent theoretical results in manifold learning.

• Nonlinear manifold learning outperforms linear methods and baseline classifiers.

• Explicitly enforcing good generalizability in the learning gives promising results.

摘要

•A nonlinear supervised dimensionality reduction method is proposed.•The learning objective is based on recent theoretical results in manifold learning.•Nonlinear manifold learning outperforms linear methods and baseline classifiers.•Explicitly enforcing good generalizability in the learning gives promising results.

论文关键词：Manifold learning,Dimensionality reduction,Supervised learning,Out-of-sample,Nonlinear embeddings

论文评审过程：Received 19 October 2017, Revised 16 August 2018, Accepted 8 October 2018, Available online 10 October 2018, Version of Record 16 October 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.10.006